Skip to content
ChatGPT · Grok · GPT-5.5Episode 108 · 3 May 2026 · 52:14

ChatGPT-5.5 Reads Prompts and Gets More Expensive: Personalization Is Growing Faster Than Trust

What to watch for

1Compare “ChatGPT as a Google replacement: conversions, interactive answers, and new interfaces” with “Image generation in everyday life: how AI helped choose a haircut”: they provide different criteria for judging the same issue.
2Test the conclusion from “The downsides of ChatGPT-5.5” in your own use case—what actually changes in the process and what remains a promise.
3Before choosing a product or approach, record the constraint identified in “ChatGPT as a Google replacement: conversions, interactive answers, and new interfaces”.
4Define the owner of the outcome and the quality metric for the situation described in ““Families of the Canada shooting victims filed a lawsuit against OpenAI and Sam Altman."”.
Signals to track afterwards
Watch for actions by Anthropic and Google that confirm or challenge the episode’s central claims.
Compare new launches and policy changes with “The downsides of ChatGPT-5.5”: have access, quality, price, or constraints changed?
Check whether the scenario in ““Families of the Canada shooting victims filed a lawsuit against OpenAI and Sam Altman."” becomes repeatable practice rather than a one-off demonstration.
Most useful for
AI usersProduct teamsEntrepreneursExecutives and managersInvestorsCompany leaders

Key takeaways

00:00Why an announcement is not enough: today in the ToTheMoon episode

The boundary of the “ChatGPT-5.5 Reads Prompts and Gets More Expensive: Personalization Is Growing Faster Than Trust” case is defined by this point: AI already makes decisions from data people consider private, and personalization can even pick a haircut, but trust cannot be built on convenience — the user has to know what is saved, who sees it, and why the model proposes exactly this.

09:03The market tests it through use: what's new in ChatGPT-5.5 — where AI got stronger

The “What's new in ChatGPT-5.5: where AI got stronger, part 2/3” scene leads to a working conclusion: the model improved on the tests OpenAI chose for its presentation, but in real work progress is less clear-cut: better phrasing can sit next to an annoying interface or price, and once again a benchmark is not a product.

12:45The boundary between value and constraint: the downsides of ChatGPT-5.5

The discussion of “The downsides of ChatGPT-5.5” yields a practical test: the downsides show precisely in real use — price, interface, unexpected behavior — which a presentation benchmark does not reveal, so judge by the workflow, not the demo.

13:34ChatGPT is replacing Google in simple actions more and more often: it converts, builds an interactive answer, explains, and proposes the next step

The “ChatGPT as a Google replacement: conversions, interactive answers, and new interfaces” topic becomes clearer once this point is included: this is convenient because the person receives a solution rather than a link. But they see the source less clearly and know less about which data the system used.

17:10What changes in real work: families of the Canada shooting victims sue OpenAI

The “Families of the Canada shooting victims file a lawsuit against OpenAI and Sam Altman” issue should be assessed with one constraint in mind: if a conversation with a model influences a person's behavior, the company cannot indefinitely treat itself as a neutral text provider, so the issue turns on responsibility and whether it can be enforced.

18:55Why context matters more than one metric: privacy of AI: Your requests are read

The “Privacy of AI: Your requests are read” topic becomes clearer once this point is included: some prompts are read by people for safety and quality, and enterprise-contract terms do not always mean absolute secrecy, so it matters to know what is retained and who sees it.

27:37How the issue moves from news to product: the Anthropic experiment — AI runs economic deals on a person's behalf

The discussion of “Anthropic experiment: AI runs economic deals on a person's behalf” yields a practical test: the model reacts to export restrictions, prices, and the other side's actions — a useful laboratory, but a real market will add manipulation, incomplete data, and legal consequences.

49:48Enterprise AI will consist of several Codex, Gemini, and Claude agents

The decision in “Image generation in everyday life: how AI helped choose a haircut” depends on one criterion: personalization can even help choose a haircut from a photo, but trust is not built on convenience: the user has to know which prompts are retained, who sees them, and why the model proposes this particular decision.

What this episode is about

The new model scores better on tests, replaces part of Google, and creates interactive answers, but users notice restrictions and price. Lawsuits, human review of prompts, and Anthropic’s experiments with economic agents show that AI is already making decisions from data people consider private.

ChatGPT-5.5 improved on the tests OpenAI chose for its presentation. In real work, progress is less clear-cut: the model may phrase an answer better while frustrating the user with the interface, price, or unexpected behavior. Once again, a benchmark is not a product.

ChatGPT is replacing Google in simple actions more and more often: it converts, builds an interactive answer, explains, and proposes the next step. This is convenient because the person receives a solution rather than a link. But they see the source less clearly and know less about which data the system used.

Lawsuits by victims’ families and other cases intensify the question of responsibility. If a conversation with a model influences human behavior, the company cannot indefinitely treat itself as a neutral provider of text. At the same time, some prompts are read by people for safety and quality, and enterprise-contract terms do not always mean absolute secrecy.

Anthropic’s experiment in which agents conduct economic transactions shows the next level. The model reacts to export restrictions, prices, and the other side’s actions. This is a useful laboratory, but a real market will add manipulation, incomplete information, and legal consequences.

Enterprise AI will consist of several Codex, Gemini, and Claude agents. Their cost grows with quality and the volume of work. Personalization can even help choose a haircut from a photograph, but trust cannot be built on convenience. Users need to know which prompts are retained, who sees them, and why the model proposes a particular decision.

Personalization can even help choose a haircut from a photograph, but trust will not be built on convenience. As a result, users need to know which prompts are retained, who sees them, and why the model proposes a particular decision.

Episode transcript

The episode is in Russian; below is an English reading guide to the transcript (the full EN transcript is a machine translation). Voice matching applied to 63 segments: 40 identified, 3 mixed, 12 probable, and 8 unresolved.

Loading…